U.S. patent application number 10/124063 was filed with the patent office on 2002-11-21 for image processing apparatus for and method of improving an image and an image display apparatus comprising the image processing apparatus.
Invention is credited to Nicolas, Marina Marie Pierre.
Application Number | 20020172420 10/124063 |
Document ID | / |
Family ID | 8180183 |
Filed Date | 2002-11-21 |
United States Patent
Application |
20020172420 |
Kind Code |
A1 |
Nicolas, Marina Marie
Pierre |
November 21, 2002 |
Image processing apparatus for and method of improving an image and
an image display apparatus comprising the image processing
apparatus
Abstract
An image processing apparatus (100) for improving images
depending on the type of regions within the image, comprises a
segmentation unit (102) to localize regions with a pre-selected
type and an improvement unit (106) being designed to improve the
image, which can be controlled in order to modulate the amount of
improvement over the image. The image processing apparatus (100)
further comprises a determination unit to find a range of values to
characterize the pixels of the regions which should be handled
separately based on the results of the segmentation unit (102).
Inventors: |
Nicolas, Marina Marie Pierre;
(Le Fontanil, FR) |
Correspondence
Address: |
U.S. Philips Corporation
580 White Plains Road
Tarrytown
NY
10591
US
|
Family ID: |
8180183 |
Appl. No.: |
10/124063 |
Filed: |
April 17, 2002 |
Current U.S.
Class: |
382/170 ;
382/173 |
Current CPC
Class: |
G06T 5/002 20130101;
G06T 2207/20192 20130101; G06T 2207/20012 20130101; G06T 5/003
20130101 |
Class at
Publication: |
382/170 ;
382/173 |
International
Class: |
G06K 009/00 |
Foreign Application Data
Date |
Code |
Application Number |
Apr 20, 2001 |
EP |
01201436.1 |
Claims
1. An image processing apparatus (100) for improving an image,
which image comprises a plurality of pixels, and the image
processing apparatus (100) comprising: a segmentation unit (102)
that is designed to perform: a first test whether a first value
(230) of a first property of a selected pixel (222) is
substantially equal to a second value (228) of a first
predetermined range (212) of values of the first property; and a
second test whether a third value (226) of a second property of the
selected pixel (222) is substantially equal to a fourth value (224)
of a second predetermined range (214) of values of the second
property, in order to segment the image into multiple regions of
pixels, each region with a region type; and an improvement unit
(106) that is designed to improve the image by processing a
particular pixel, depending on the region type of the region to
which the particular pixel corresponds, characterized in comprising
a determination unit (104) that is designed to characterize pixels
corresponding to regions with a pre-determined region type by
determination of an actual range (216) of values of a particular
property of the pixels, other than a location; and that the
improvement unit (106) processes the pixels corresponding to
regions with the pre-determined region type differently.
2. An image processing apparatus (100) as claimed in claim 1,
characterized in that the improvement unit (106) is designed to
reduce noise in the image.
3. An image processing apparatus (100) as claimed in claim 1,
characterized in that the improvement unit (106) is designed to
enhance details in the image.
4. An image processing apparatus (100) as claimed in claim 1,
characterized in that the first predetermined range (212) of
possible values comprises the values related to luminance.
5. An image processing apparatus (100) as claimed in claim 1,
characterized in that the first predetermined range (212) of
possible values comprises the values related to color.
6. An image processing apparatus (100) as claimed in claim 1,
characterized in that the first test is to verify whether a first
difference between the first value (230) of the first property of
the selected pixel (222) and a sixth value of the neighboring pixel
(223) is substantially equal to a seventh value of a third
predetermined range of values of the first property.
7. An image processing apparatus (100) as claimed in claim 1,
characterized in comprising: a histogram generating means (114) for
generating a histogram (116) of values of the particular property,
whereby the histogram comprises pixels for which the first test and
the second test are positive; and an analyzer (118) designed to
analyze the histogram (116) in order to determine a classification
value of the pixels of the image based on the histogram (116), with
the classification value to control the amount of improvement.
8. An image processing apparatus (100) as claimed in claim 7,
characterized in that a maximum classification value (418)
corresponds to a center of the actual range (216), and a minimum
classification value (420) corresponds to a border of the actual
range (216), the classification value varying substantially
continuously from the center of the actual range (216) towards the
border of the actual range.
9. An image processing apparatus (100) as claimed in claim 1,
characterized in that the actual range (216) of values comprises
the values related to color.
10. An image processing apparatus (100) as claimed in claim 1,
characterized in that the actual range (216) of values comprises
the values related to luminance.
11. An image processing apparatus (100) as claimed in claim 1,
characterized in that the actual range (216) of values comprises
values related to a second difference between the particular value
of the particular property of the selected pixel (222) and a ninth
value of the particular property of the neighboring pixel
(223).
12. An image display apparatus (300) provided with: receiving means
(302) for receiving a video signal representing images; an image
processing apparatus (100) for improving an image, which image
comprises a plurality of pixels, and the image processing apparatus
(100) comprising: a segmentation unit (102) that is designed to
perform: a first test whether a first value (230) of a first
property of a selected pixel (222) is substantially equal to a
second value (228) of a first predetermined range (212) of values
of the first property; and a second test whether a third value
(226) of a second property of the selected pixel (222) is
substantially equal to a fourth value (224) of a second
predetermined range (214) of values of the second property, in
order to segment the image into multiple regions of pixels, each
region with a region type; and an improvement unit (106) that is
designed to improve the image by processing a particular pixel,
depending on the region type of the region to which the particular
pixel corresponds; and a display device (306) for displaying the
images, characterized in that the image processing apparatus (100)
comprises a determination unit (104) that is designed to
characterize pixels corresponding to regions with a predetermined
region type by determination of an actual range (216) of values of
a particular property of the pixels, other than a location; and
that the improvement unit (106) processes the pixels corresponding
to regions with the pre-determined region type differently.
13. A method of improving an image, which image comprises a
plurality of pixels, and the image processing apparatus (100)
comprising: a segmentation step to perform: a first test whether a
first value (230) of a first property of a selected pixel (222) is
substantially equal to a second value (228) of a first
predetermined range (212) of values of the first property; and a
second test whether a third value (226) of a second property of the
selected pixel (222) is substantially equal to a fourth value (224)
of a second predetermined range (214) of values of the second
property, in order to segment the image into multiple regions of
pixels, each region with a region type; and an improvement step to
improve the image by processing a particular pixel, depending on
the region type of the region to which the particular pixel
corresponds, characterized in comprising a determination step to
characterize pixels corresponding to regions with a predetermined
region type by determination of an actual range (216) of values of
a particular property of the pixels, other than a location; and
that the improvement unit (106) processes the pixels corresponding
to regions with the pre-determined region type differently.
Description
[0001] The invention relates to an image processing apparatus for
improving an image, which image comprises a plurality of pixels,
and the image processing apparatus comprising:
[0002] a segmentation unit that is designed to perform:
[0003] a first test whether a first value of a first property of a
selected pixel is substantially equal to a second value of a first
predetermined range of values of the first property; and
[0004] a second test whether a third value of a second property of
the selected pixel is substantially equal to a fourth value of a
second predetermined range of values of the second property, in
order to segment the image into multiple regions of pixels, each
region with a region type; and
[0005] an improvement unit that is designed to improve the image by
processing a particular pixel, depending on the region type of the
region to which the particular pixel corresponds.
[0006] The invention further relates to an image display apparatus
provided with:
[0007] receiving means for receiving a video signal representing
images;
[0008] such an image processing apparatus;
[0009] a display device for displaying the images.
[0010] The invention further relates to a method of improving an
image, which image comprises a plurality of pixels, the method
comprising:
[0011] a segmentation step to perform:
[0012] a first test, being a check whether a first value of a first
property of a selected pixel is substantially equal to a second
value of a first predetermined range of values of the first
property; and
[0013] a second test, being a check whether a third value of a
second property of the selected pixel is substantially equal to a
fourth value of a second predetermined range of values of the
second property, in order to segment the image into multiple
regions of pixels, each region with a region type; and
[0014] an improvement step to improve the image to improve the
image by processing a particular pixel, depending on the region
type of the region to which the particular pixel corresponds.
[0015] An image processing apparatus and a method of the kind
described in the opening paragraph are disclosed in U.S. Pat. No.
5,848,181.
[0016] The known apparatus provides location dependent noise
reduction. The known apparatus varies the amount of noise reduction
over the image: a relatively high noise reduction in some regions
and a lower noise reduction in other regions. This modulation is to
avoid that details are removed by noise reduction. The level of
noise reduction required in flat, low detailed, regions is often
too high for detailed regions of the same image. Such a high level
of noise reduction might result in loss of information. It might
also lead to images looking unnatural. There is no reason why the
amount of white noise would differ from one region of the image to
the other. However, a particular amount of noise is more visible in
a flat region of the image than in another, detailed region. In
order to achieve an appropriate location dependent noise reduction,
a kind of segmentation or classification is required to divide an
image into regions which are relatively flat respectively in
regions which are more crispy. In the known image processing
apparatus the segmentation of the image into regions is based on
differences in luminance values of neighboring pixels. The
differences in luminance values are used to determine
pixel-position dependent coefficients. The noise reduction is then
modulated taking into account the pixel-position dependent
coefficients. A disadvantage of this known apparatus is that it is
very difficult, or even impossible to implement it for real time
applications without making use of an image memory and optionally
complicated motion-compensation algorithms, which makes it costly.
Another disadvantage is that the known apparatus is very sensitive
to the segmentation precision with the risk of having artifacts at
object borders. E.g. suppose there is an image comprising grass
with some small flowers in it. From a segmentation point of view it
might be that one grass region is detected. But the isolated pixels
corresponding to the small flowers are not segmented separately
although they have conditions different from the typical grass
conditions. In this case, green enhancement might be required to
the surrounding pixels, but not to the isolated pixels.
[0017] It is a first object of the invention to provide an image
processing apparatus of the kind described in the opening paragraph
which is arranged to perform non-position controlled image
improvement.
[0018] It is a second object of the invention to provide an image
display apparatus of the kind described in the opening paragraph
comprising an image processing apparatus in which the improvement
is not position controlled.
[0019] It is a third object of the invention to provide a method of
the kind described in the opening paragraph in which the
improvement is not position controlled.
[0020] The first object of the invention is achieved in that the
image processing apparatus comprises a determination unit that is
designed to characterize pixels corresponding to regions with a
pre-determined region type by determination of an actual range of
values of a particular property of the pixels, other than a
location; and that the improvement unit processes the pixels
corresponding to regions with the pre-determined region type
differently. The determination unit is designed to determine for
the image under consideration the actual range of values of a
property of the pixels which should be handled separately. The
particular property may be equal to the first or second property.
The determination unit allows to de-couple the segmentation unit
from the improvement unit.
[0021] For instance the segmentation unit can be placed basically
at any place in the processing chain, provided that no improvement
feature, placed before it will significantly affect the properties
considered for the pre-determined region type. By de-coupling it is
not meant that the segmentation unit is completely independent from
the improvement unit. The segmentation unit in combination with the
determination unit identifies conditions, or range of values, that
will require a specific setting in the improvement unit, rather
than a specific object. Basically, the improvement unit does not
need to know what the picture is representing, but the conditions
for which a separate setting is required to avoid non-optimal
performance.
[0022] A major advantage of the de-coupling between segmentation
unit and the improvement unit caused by the determination unit is
that it enables a relatively easy real-time implementation of the
image processing apparatus. In the case of a sequence of images to
be processed use can be made of the temporal continuity of the
video signal representing the sequence of images. By temporal
continuity it is meant that the conditions that will characterize
one region will hardly change from one image to the other. In that
case a segmentation can be performed on an image and the results of
this segmentation can be used, after translation to condition data,
without performing motion compensation, in order to apply an
improvement on a subsequent image. In stead of images, i.e. video
frames also video fields can be taken as entities for this
substitution. By using coordinates to indicate pixels belonging to
a region, this is hardly possible, because it is very likely that
an object which is computed at pixel position (x,y) within the
image as present in the segmentation unit has moved to another
position in the improvement unit. For instance, grass pixels might
move from one location in the first image to another location in
the subsequent image, but it is very likely that the conditions
that make them grass in one image, e.g. the value of the color and
the luminance gradient around the pixel, will remain nearly
unchanged in the next image. Therefor the image processing
apparatus does not require precise motion compensation. Another
advantage is that the location of the image processing apparatus
within the chain of processing units of an image display apparatus
is less dependent of the architecture of the image display
apparatus. It is even possible that the segmentation unit and the
improvement unit are allocated in different parts of the image
display apparatus. E.g. the segmentation unit in the begin of the
processing chain, followed by some other processing units and then
the improvement unit or with improvement units spread over the
whole video path. A major other advantage of the system is that,
while the segmentation can be made rather sophisticated, the
conditions transmitted to the improvement units can be most of the
time heavily simplified. For instance, to identify grass, the
segmentation unit needs to take into account color and texture
information. But, for a given image, if it turns out, at the output
of the determination unit that all the green pixels in the picture
do have the texture of grass, then the improvement unit only needs
to check whether the pixel is green or not.
[0023] In an embodiment of the image processing apparatus according
to the invention the improvement unit is designed to reduce noise
in the image. The rationale for regional noise reduction is
explained: to avoid that details are removed by noise reduction and
to remove sufficiently noise in flat regions. This enables to
modulate the amount of noise reduction over regions of the image.
Several types of noise reduction can be realized: e.g. spatial,
temporal and spatio-temporal. For a temporal or spatio-temporal
noise reduction a sequence of images is required. For both the
classification and the noise reduction it is possible to use more
than one image.
[0024] In an embodiment of the image processing apparatus according
to the invention the improvement unit is designed to enhance
details in the image. Details may be edges or textures.
[0025] In an embodiment of the image processing apparatus according
to the invention the first predetermined range of values comprises
the values related to color. Knowledge of the scene which has been
imaged, makes it possible to extract describing parameters. In the
case that it is known that e.g. a football match has been imaged,
then it is quite certain that grass will be visible in the images.
Because the predetermined range of color values of the image
processing apparatus can correspond to the colors of grass this
embodiment can be tuned to specific types of scenes. Embodiments
with another predetermined range of color values can be tuned to
other specific types of scenes, e.g. blue sky, blue water, brown
sand, white snow or flesh-tone for human skin.
[0026] In an embodiment of the image processing apparatus according
to the invention the first test is to verify whether a difference
between a first value of the first property of the selected pixel
and a ninth value of the neighboring pixel is substantially equal
to a value of a third predetermined range of values of the first
property. E.g. differences in luminance values are related to the
spectrum of frequency components of the image. This spectrum offers
a lot of information about noise and hence can be applied to detect
noise.
[0027] An embodiment of the image processing apparatus according to
the invention comprises:
[0028] a histogram generating means for generating a histogram of
values of the particular property, whereby the histogram comprises
pixels for which the first test and the second test are
positive;
[0029] an analyzer designed to analyze the histogram in order to
determine a classification value of the pixels of the image based
on the histogram, with the classification value to control the
amount of improvement.
[0030] The classification values are determined by analyzing the
properties of the histogram:
[0031] If the histogram is relatively narrow and symmetrical, then
the probability that the pixels of the histogram correspond to a
region with a pre-selected region type is relatively high;
[0032] If the histogram is relatively spread over a large range of
values or is unsymmetrical, then the probability that the pixels of
the histogram correspond to a region with a pre-selected region
type is relatively low.
[0033] After having analyzed the histogram the analyzer classifies
the pixels. This classification resembles the probability that the
pixels of the image meeting the conditions of the tests, correspond
to a region with a pre-selected region type. Pixels with mutually
substantially equal values get the same classification value. By
varying the classification values the amount of image improvement
can not only be controlled region by region but even with a lower
granularity.
[0034] In an embodiment of the image processing apparatus according
to the invention comprising the histogram generating means, a
maximum classification value corresponds to a center of the actual
range, and a minimum classification value corresponds to a border
of the actual range, the classification value varying substantially
continuously from the center of the actual range towards the border
of the actual range. This classification "profile" is based on the
following assumptions:
[0035] The probability that a pixel at the center of a region has a
value which is substantially equal to the value at the center of
the actual range is maximum;
[0036] The probability that a pixel at a border of a region has a
value which is substantially equal to the value at a border of the
actual range is minimum;
[0037] When moving from the center of a region to the border of the
region, the luminance, color and basic gradient values also change
monotonously and quite continuously from typical values to border
values.
[0038] The second object of the invention is achieved in that the
image display apparatus comprises an image processing apparatus
with a determination unit that is designed to characterize pixels
corresponding to regions with a pre-determined region type by
determination of an actual range of values of a particular property
of the pixels, other than a location; and that the improvement unit
processes the pixels corresponding to regions with the
pre-determined region type differently.
[0039] The third object of the invention is achieved in that the
method comprises a determination step to characterize pixels
corresponding to regions with a pre-determined region type by
determination of an actual range of values of a particular property
of the pixels, other than a location; and that the improvement unit
processes the pixels corresponding to regions with the
pre-determined region type differently.
[0040] These and other aspects of the image processing apparatus,
the method of improving an image and of the image display apparatus
according to the invention will become apparent from and will be
elucidated with reference with respect to the implementations and
embodiments described hereinafter and with reference to the
accompanying drawings, wherein:
[0041] FIG. 1 schematically shows elements of the image processing
apparatus;
[0042] FIG. 2A schematically shows an image with two regions;
[0043] FIG. 2B schematically shows a probability space;
[0044] FIG. 3 schematically shows elements of the image display
apparatus;
[0045] FIG. 4A schematically shows a probability space;
[0046] FIG. 4B schematically shows a histogram of pixels; and
[0047] FIG. 4C schematically shows a classification function.
[0048] FIG. 1 schematically shows the following elements of the
image processing apparatus 100:
[0049] a segmentation unit 102 that is designed to perform a number
of tests in order to segment the image into multiple regions of
pixels, each region with a region type;
[0050] a determination unit 104 that is designed to determine an
actual range of values, in order to be able to indicate the region
type of the region to which the selected pixel corresponds;
[0051] an improvement unit 106 that is designed to improve the
image by processing the pixels, depending on the region type of the
region to which the pixels correspond;
[0052] a histogram generating means 114 for generating a histogram
116 of values of the particular property, whereby the histogram
comprises pixels for which the tests are positive; and
[0053] an analyzer 118 designed to analyze the histogram 116 in
order to determine a classification value of the pixels of the
image based on the histogram 116.
[0054] The image enters the image processing apparatus 100 at the
input connector 112. Improved image is provided at the output
connector 110. The predetermined ranges of values for the tests in
the segmentation unit 102 can be adjusted by means of the control
input 126. The segmentation unit 102 may use simple pixel-based
tests, but also much more complicated ones like block-based, i.e.
texture related tests are possible. The output of the segmentation
unit contains either condition information directly or, if e.g.
concealment is used in the segmentation, location information
P(x,y) which represents the probability of the pixel at location
(x,y) to belong to a particular region. However the goal of the
segmentation unit 102 in combination with determination unit 104 is
not to identify the location of objects, but to determine whether
there are regions in the image where global settings of the image
improvement will not be adequate and to find at least one range of
values to characterize these regions. For instance, an apparatus
for noise reduction in images with optionally grass in it, is not
interested in identifying grass, but just determines the
probability that a pixel belongs to. Artifacts of noise reduction
are the most annoying in some grass textures. The segmentation in
that case does not look for every kind of grass but for problematic
grass. To detect grass there is more required than just detect
green. But if it turns out, e.g. by inspecting the histogram in
116, that the grass region of the image is the only place where
green is present, then the only information that must be
transmitted to the improvement unit 106 is whether a pixel is green
or not. The spatial-based control is then translated to a simple
condition-based control.
[0055] Another example of a location dependent image improvement
apparatus is an apparatus being able to discriminate between
foreground and background. This will be explained by means of FIG.
2A and FIG. 2B. FIG. 2A shows an image 200 with a foreground region
204 and background region 202. A selected pixel 222 is indicated.
The selected pixel 222 is also depicted in FIG. 2B. FIG. 2B shows a
probability space 206 in which the pixels of image 200 are put. The
horizontal axis 208 resembles a first property of the pixels of the
image 200. The vertical axis 210 resembles a second property of the
pixels of the image 200. The value of the first property of the
selected pixel 222 is referenced with 230. The value of the second
property of the selected pixel 222 is referenced with 226. Suppose
it is required to segment the image into a foreground region 204
and a background region 202. In order to segment the image a number
of test are applied. For a first test the differences between
luminance values of neighboring pixels are calculated and checked
whether this difference is in the range of possible values of the
first property 212. This means that in this case the first property
corresponds to the difference between luminance values of
neighboring pixels. For a second test the color value of the pixels
is checked. For the selected pixel 222 this second test is a
verification whether the value 226 is in the range of possible
values of the second property 214. Because the probability to find
background 202 in the upper portion of the image 200 is relatively
high, the location of the pixels in the image 200 are also used for
the segmentation. After performing the tests for all pixels of the
image 200 it appears that two sets of pixels can be distinguished
in the probability space 206: a first set of pixels 218 and a
second set of pixels 220. The first set of pixels 218 can be
characterized by using a range of actual values 216 on the axis of
the second property. In other words it appears that background
pixels and foreground pixels can be separated by means of their
color value, for this particular image.
[0056] FIG. 3 shows elements of an image display apparatus 300
according to the invention. The image display apparatus 300 has a
receiving means 302 for receiving a video signal representing the
images to be displayed. The signal may be a broadcast signal
received via an antenna or cable but may also be a signal from a
storage device like a VCR (Video Cassette Recorder) or DVD (Digital
Versatile Disk). The image display apparatus 300 further has an
image processing apparatus 100 for processing the video signal and
a display device 306 for displaying the images represented by the
improved video signal. The image processing apparatus 100 is
implemented as described in FIG. 1. The image processing apparatus
100 can be controlled externally. The predetermined ranges of
values for the tests in the segmentation unit can be adjusted by
means of the control input 126. The receiving means 302 can get
notified or has capabilities to extract information from the video
signal about the type of scene that has been imaged in order to set
the appropriate ranges of predetermined values of the image display
apparatus 300. E.g. in the case of a football match a predetermined
range of colors should match the colors of grass.
[0057] FIGS. 4A, 4B and 4C depict the relation between the actual
ranges of values and classification values. FIG. 4A shows a
probability space in which the pixels of an image are put after
performing a number of tests. The horizontal axis 406 resembles a
first property of the pixels of the image. The vertical axis 408
resembles a second property of the pixels of the image. It appears
that two sets of pixels can be distinguished in the probability
space: a first set of pixels 402 and a second set of pixels 404.
The first set of pixels 402 can be characterized by using a range
of actual values 216 on the axis of a particular property. A
histogram 116 is made of pixels for which the tests were positive.
FIG. 4B shows this histogram. The horizontal axis 412 resembles the
particular property of the pixels of the image. The vertical axis
410 resembles the number of pixels of the image that satisfy the
tests and have a pre-selected value of the particular property.
FIG. 4C shows a classification function used to classify the
pixels. The classification function shows the classification value
as function of value of the particular property. The horizontal
axis 416 resembles the particular property of the pixels of the
image. The vertical axis 414 resembles the classification value.
The maximum classification value 418 corresponds to the center of
the actual range 216. The minimum classification value 420
corresponds to the borders of the actual range 216. The
classification value varies substantially continuously from the
center of the actual range 216 towards the borders of the actual
range.
[0058] It should be noted that the above-mentioned embodiments
illustrate rather than limit the invention and that those skilled
in the art will be able to design alternative embodiments without
departing from the scope of the appended claims. In the claims, any
reference signs placed between parentheses shall not be constructed
as limiting the claim. The word `comprising` does not exclude the
presence of elements or steps not listed in a claim. The word "a"
or "an" preceding an element does not exclude the presence of a
plurality of such elements. The invention can be implemented by
means of hardware comprising several distinct elements and by means
of a suitable programmed computer. In the unit claims enumerating
several means, several of these means can be embodied by one and
the same item of hardware.
* * * * *